Fault diagnosis model training method, fault diagnosis method and related equipment

By introducing an ELM model on the preliminary diagnosis neural network, extracting and training the distribution weights of composite faults and single faults, and building a fault diagnosis model, the problem of complex fault diagnosis is solved and the recognition accuracy is improved.

CN119066601APending Publication Date: 2024-12-03HANGZHOU ANMAISHENG INTELLIGENT TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310627868.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-30
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Compound faults occur simultaneously on the device, resulting in increased diagnostic difficulty. The existing artificial neural network-based methods have low recognition accuracy when there are fewer samples.

Method used

On the basis of the preliminary diagnosis neural network, the ELM model was introduced. By extracting the characteristics of single fault and fault-free data, the preliminary diagnosis neural network was trained, the composite fault distribution weight and single fault distribution weight were obtained, and the composite fault distribution weight was input to the ELM model for further training. Build a fault diagnosis model.

Benefits of technology

This improves the accuracy of composite fault diagnosis, avoids the problem of low recognition accuracy of preliminary diagnosis neural networks when there are fewer composite fault samples, and enhances the recognition accuracy of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066601A_ABST
    Figure CN119066601A_ABST
Patent Text Reader

Abstract

The invention discloses a fault diagnosis model training method, a fault diagnosis method and related equipment, which are applied to the field of fault diagnosis, and the method comprises the steps: extracting a first feature of single fault data and a second feature of fault-free data, and training a preliminary diagnosis neural network according to the first feature and the second feature; extracting a third feature of the composite fault data and inputting the third feature into the preliminary diagnosis neural network to obtain a composite fault distribution weight; inputting the first feature into a preliminary diagnosis neural network to obtain a single fault distribution weight; training an ELM model according to the composite fault distribution weight and the single fault distribution weight; and connecting the output layer of the preliminary diagnosis neural network with the input layer of the ELM model to obtain a fault diagnosis model. According to the method, the ELM model is added for further detection on the basis of diagnosis by the preliminary diagnosis neural network, so that the problem of low accuracy of a diagnosis result obtained by performing composite fault detection by using the preliminary diagnosis neural network under the condition of few composite fault samples is avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a fault diagnosis model training method, a fault diagnosis method, a fault diagnosis model training device, a fault diagnosis device, equipment and a computer-readable storage medium. Background Art

[0002] Compound faults refer to situations where multiple faults occur simultaneously on the same equipment. It may be that the aging of the equipment causes the degradation of different parts at the same time, or it may be that the vibration, temperature rise and other phenomena caused by a fault induce more faults. In the actual production process, if the possible abnormal causes of the equipment are not promptly analyzed and investigated and relevant measures are not taken in time, production may stagnate, resulting in economic and human losses and waste. There are many reasons for the formation of compound faults, and the combination of faults is changeable. In addition, the fault characteristics of multiple faults will overlap and interfere with each other. Therefore, it is difficult to diagnose compound faults, and it is necessary to accurately distinguish and identify the different faults that constitute it.

[0003] With the development and application of machine learning and artificial intelligence algorithms, there are many studies and applications of equipment fault diagnosis methods based on intelligent algorithms. Since general artificial neural network methods require a large amount of data to train models, and the situations of compound faults are changeable and the samples are relatively small, the recognition accuracy of the trained models is low. Summary of the invention

[0004] The purpose of the present invention is to provide a fault diagnosis model training method, a fault diagnosis method and related equipment, which are applied to the field of fault diagnosis. The method adds an ELM model for further detection on the basis of a preliminary diagnosis neural network to avoid the problem of low accuracy of the diagnosis results obtained by using a preliminary diagnosis neural network for composite fault detection when there are fewer composite fault samples.

[0005] In order to solve the above technical problems, the present invention provides a fault diagnosis model training method, comprising:

[0006] Extracting a first feature of single fault data and a second feature of no-fault data, and inputting the first feature and the second feature into a preliminary diagnosis neural network for model training;

[0007] Extracting a third feature of the composite fault data, and inputting the third feature into the trained preliminary diagnosis neural network to obtain a composite fault distribution weight;

[0008] Inputting the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight;

[0009] Inputting the composite fault distribution weight and the single fault distribution weight into an ELM model for training;

[0010] The output layer of the trained preliminary diagnosis neural network is connected to the input layer of the trained ELM model to obtain a fault diagnosis model.

[0011] Optionally, the extracting the first feature of the single fault data and the second feature of the no-fault data includes:

[0012] Processing the single fault vibration data by wavelet packet decomposition and energy spectrum calculation to obtain the first feature;

[0013] Processing the fault-free vibration data by the wavelet packet decomposition and the energy spectrum calculation to obtain the second feature;

[0014] Accordingly, the third feature of the composite fault data is extracted, including:

[0015] The third feature is obtained by processing the composite fault vibration data through the wavelet packet decomposition and the energy spectrum calculation.

[0016] Optionally, the wavelet packet decomposition is a four-layer wavelet packet decomposition.

[0017] Optionally, before extracting the first feature of the single fault data and the second feature of the no-fault data, the method further includes:

[0018] The single fault data and the non-fault data are processed by resampling and downsampling, so that the ratio of the amount of each fault type data in the single fault data to the amount of non-fault data is equal.

[0019] Optionally, the preliminary diagnosis neural network is a BP neural network.

[0020] In order to solve the above technical problems, the present invention also provides a fault diagnosis method, comprising:

[0021] Acquire the characteristics of the data to be diagnosed;

[0022] Inputting the features into a fault diagnosis model to obtain a fault type diagnosis result;

[0023] Wherein, the fault diagnosis model is a model trained according to any of the above-mentioned fault diagnosis model training methods.

[0024] In order to solve the above technical problems, the present invention also provides a fault diagnosis model training device, comprising:

[0025] A first training module is used to extract a first feature of single fault data and a second feature of no-fault data, and input the first feature and the second feature into a preliminary diagnosis neural network for model training;

[0026] A first weight module is used to extract a third feature of the composite fault data, and input the third feature into the trained preliminary diagnosis neural network to obtain a composite fault distribution weight;

[0027] A second weight module, used for inputting the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight;

[0028] A second training module, used for inputting the composite fault distribution weight and the single fault distribution weight into an ELM model for training;

[0029] The model building module is used to connect the output layer of the trained preliminary diagnosis neural network with the input layer of the trained ELM model to obtain a fault diagnosis model.

[0030] In order to solve the above technical problems, the present invention further provides a fault diagnosis device, comprising:

[0031] An acquisition module, used to acquire the characteristics of the data to be diagnosed;

[0032] An output module, used for inputting the features into a fault diagnosis model to obtain a fault type diagnosis result;

[0033] Wherein, the fault diagnosis model is a model trained according to the above-mentioned fault diagnosis model training device.

[0034] In order to solve the above technical problems, the present invention further provides a fault diagnosis device, comprising:

[0035] Memory, for storing computer programs;

[0036] A processor is used to implement any of the above-mentioned fault diagnosis model training methods and / or the above-mentioned fault diagnosis methods when executing the computer program.

[0037] In order to solve the above technical problems, the present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, any one of the above-mentioned fault diagnosis model training methods and / or the above-mentioned fault diagnosis methods is implemented.

[0038] It can be seen that the present invention extracts the first feature of single fault data and the second feature of no-fault data, inputs the first feature and the second feature into the preliminary diagnosis neural network for model training; extracts the third feature of compound fault data, inputs the third feature into the trained preliminary diagnosis neural network to obtain the compound fault distribution weight; inputs the first feature into the trained preliminary diagnosis neural network to obtain the single fault distribution weight; inputs the compound fault distribution weight and the single fault distribution weight into the ELM model for training; connects the output layer of the trained preliminary diagnosis neural network with the input layer of the trained ELM model to obtain a fault diagnosis model. And obtain the features of the data to be diagnosed; input the features into the fault diagnosis model to obtain the fault type diagnosis result. This method avoids the problem of low accuracy of the diagnosis result obtained by using the preliminary diagnosis neural network for compound fault detection when there are few compound fault samples by adding the ELM model for further detection on the basis of the diagnosis by the preliminary diagnosis neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A flowchart of a fault diagnosis model training method provided by an embodiment of the present invention;

[0041] Figure 2 A flowchart of a fault diagnosis method provided by an embodiment of the present invention;

[0042] Figure 3 An example diagram of frequency spectrum characteristics of unbalanced fault data provided by an embodiment of the present invention;

[0043] Figure 4 An example diagram of frequency spectrum characteristics of misalignment fault data provided by an embodiment of the present invention;

[0044] Figure 5 An example diagram of frequency spectrum characteristics of composite fault data provided by an embodiment of the present invention;

[0045] Figure 6 An example diagram of the frequency spectrum characteristics of fault-free data provided by an embodiment of the present invention;

[0046] Figure 7 A structural block diagram of a fault diagnosis model training device provided by an embodiment of the present invention;

[0047] Figure 8This is a structural block diagram of a fault diagnosis device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The following combination Figure 1 , Figure 1 A flowchart of a fault diagnosis model training method provided by an embodiment of the present invention, the method may include:

[0050] S101: extracting the first feature of single fault data and the second feature of no-fault data, and inputting the first feature and the second feature into a preliminary diagnosis neural network for model training.

[0051] This embodiment first trains a preliminary diagnostic neural network through each single fault data and fault-free data. This embodiment does not limit the type of single fault data, which can be vibration data collected on the equipment. This embodiment also does not limit the way to extract the first feature and the second feature from the single fault data and fault-free data. Generally, for vibration data greater than 1000Hz, four-layer wavelet packet decomposition and energy spectrum calculation can be used to extract characteristic values. Wavelet packet is a high-frequency vibration analysis method proposed on the basis of wavelet analysis. Its principle comes from the tower algorithm of multi-scale analysis. The entire frequency band is decomposed by wavelet packet to better observe the components of each frequency band to extract fault feature quantities. This embodiment can sample the high-frequency vibration signal and divide it into 2n nodes. The sum of the squares of the signals in each frequency band after decomposition is used as the wavelet packet energy. The energy spectrum value corresponding to the xth frequency band is:

[0052]

[0053] Among them, E x represents the xth energy spectrum, N represents the length of the original signal, m = 0, 1, ..., 2 n -1, It represents the kth coefficient corresponding to the node S(n,m) after wavelet packet decomposition. The extracted features can be expressed as E=[E1,E2,…,E2 n ].

[0054] After extracting the first feature and the second feature, this embodiment can input a preliminary diagnosis model for training. This embodiment does not limit the type of the preliminary diagnosis model, and can be some commonly used neural networks, such as BP (Back Propagation) neural network. BP neural network is a multi-layer feedforward neural network based on the error back propagation algorithm. The number of intermediate layers and the number of neurons in each layer of the network can be arbitrarily set according to the specific situation. This embodiment can select BP neural network as the preliminary diagnosis neural network. Further, in this embodiment, the structure and parameters of the neural network can be set according to the actual application scenario.

[0055] Furthermore, in order to ensure the rationality of model construction, each single fault data and fault-free data can be processed by downsampling and resampling to make the ratio of the data volume of each fault type in the single fault data equal to the data volume of the fault-free data.

[0056] S102: extracting the third feature of the composite fault data, and inputting the third feature into the trained preliminary diagnosis neural network to obtain the composite fault distribution weight.

[0057] S103: Inputting the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight.

[0058] In this embodiment, the third feature of the composite fault data can be extracted. The method of extracting the third feature in this embodiment needs to correspond to the method of extracting the first feature and the second feature. After the third feature is extracted, the third feature can be input into the trained preliminary diagnosis neural network to obtain the composite fault distribution weight. And the first feature can be input into the trained preliminary diagnosis neural network to obtain the single fault distribution weight.

[0059] The single fault distribution weight and load fault weight in this embodiment can be understood as the result of preliminary diagnosis through the preliminary diagnosis neural network model. Since the general artificial neural network method requires a large amount of data to train the model, and the situation of compound faults is changeable, the samples are small, and the recognition accuracy of the trained model is low. That is, the result of fault diagnosis based on the obtained single fault distribution weight and load fault weight in this embodiment is not accurate.

[0060] S104: Inputting the composite fault distribution weight and the single fault distribution weight into the ELM model for training.

[0061] In order to solve the above problems, the present invention introduces the ELM (Extreme Learning Machine) model. Compared with conventional neural networks, the ELM extreme learning machine still has high accuracy in the case of small samples and can effectively avoid falling into local optimal solutions and slow convergence. It is suitable for constructing a small sample composite fault discrimination model.

[0062] This embodiment introduces the concept of fault type distribution weights on the basis of the original preliminary diagnosis neural network, extracts features as input for preliminary diagnosis neural network training through feature extraction algorithm, defines the output as the distribution weights of each fault type, and inputs the obtained composite fault distribution weights and single fault distribution weights into the ELM model for training, and obtains the trained ELM model for further accurate diagnosis of the preliminary diagnosis results. This embodiment does not limit the structure and parameters of the ELM model, which can be set according to the actual application scenario.

[0063] S105: Connecting the output layer of the trained preliminary diagnosis neural network to the input layer of the trained ELM model to obtain a fault diagnosis model.

[0064] In this embodiment, the output layer of the trained preliminary diagnosis neural network can be connected to the input layer of the trained ELM model to obtain a fault diagnosis model.

[0065] The embodiment of the present invention avoids the problem of low accuracy of diagnostic results obtained by using the preliminary diagnostic neural network for composite fault detection when there are few composite fault samples by adding the ELM model for further detection.

[0066] The following combination Figure 2 , Figure 2 A flowchart of a fault diagnosis method provided by an embodiment of the present invention, the method may include:

[0067] S201: Acquire features of data to be diagnosed.

[0068] S202: Input the features into the fault diagnosis model to obtain a fault type diagnosis result.

[0069] The fault diagnosis model is a model trained by any fault diagnosis model training method.

[0070] This embodiment can train a fault diagnosis model according to any fault diagnosis model training method, and this embodiment can extract the features of the data to be diagnosed, and input the features into the fault diagnosis model to obtain the fault diagnosis result. This embodiment does not limit the content of the output fault diagnosis result, which generally changes according to the different model structures. For example, in this embodiment, the fault diagnosis result can only determine whether it is a composite fault, or can output a specific single fault type and load fault type.

[0071] The embodiment of the present invention avoids the problem of low accuracy of diagnostic results obtained by using the preliminary diagnostic neural network for composite fault detection when there are few composite fault samples by adding the ELM model for further detection.

[0072] The following is a specific embodiment of a fault diagnosis method provided by an embodiment of the present invention, wherein the device verified by the specific embodiment may be a large centrifugal fan in a domestic power plant, and the types of faults verified are: imbalance, misalignment, and a combined fault of imbalance and misalignment. When training the model, the BP neural network may use [0.1, 0.9] to represent an unbalanced fault, [0.9, 0.1] to represent a misaligned fault, and [0.1, 0.1] to represent normal operation of the equipment. The specific embodiment may include:

[0073] The preliminary diagnosis neural network is set to BP neural network, and the structural parameters are shown in Table 1:

[0074] Table 1 shows the structural parameters of the BP neural network model.

[0075]

[0076] The structural parameters of the ELM model can be shown in Table 2:

[0077] Table 2 shows the structural parameters of the ELM model.

[0078]

[0079] The unbalanced sample data, misaligned sample data and fault-free sample data are processed by resampling and downsampling to make the data volume ratio equal; among which, the frequency spectrum characteristic example of the unbalanced fault data can be shown as follows Figure 3 As shown in the figure, the frequency spectrum characteristic example of misalignment fault data can be shown as Figure 4 As shown, the frequency spectrum characteristic example of composite fault data can be shown as Figure 5 As shown, the frequency spectrum characteristic example of fault-free data can be shown as Figure 6 As shown; Figures 3 to 6 In the time domain diagram, the horizontal axis represents time (unit: second), and the vertical axis represents the value (unit: mm / s); the horizontal axis of the frequency domain diagram represents frequency (unit: Hz), and the vertical axis represents the value (unit: mm / s); the horizontal axis of the energy spectrum represents the frequency interval, and the vertical axis represents the energy in each frequency interval after wavelet packet decomposition.

[0080] In the ELM model, the output uses “0” to indicate a non-compound fault and “1” to indicate a compound fault of imbalance and misalignment;

[0081] All data from the power plant from September 2020 to December 2022 were used as sample data, and 252 sets of single fault data and fault-free data were obtained. 80% of the data, a total of 201 sets, were used for BP neural network model training, and 20% of the data, a total of 51 sets, were used as test sets.

[0082] A total of 32 groups of compound fault samples were sorted out, 12 groups were selected as test sets, and 20 groups of data were selected as inputs to the trained BP neural network to obtain the compound fault distribution weights;

[0083] Input the single fault data into the trained BP neural network to obtain the single fault distribution weight;

[0084] The single fault distribution weight and the compound fault distribution weight are input into the ELM model for training, and the output layer of the trained BP neural network is connected with the input layer of the trained ELM model to obtain a fault diagnosis model;

[0085] The features of the test set data are extracted and input into the fault diagnosis model for diagnosis. The recognition success rate of 12 groups of compound faults is tested. The test results are shown in Table 3:

[0086] Table 3 shows the comparison of composite fault diagnosis results.

[0087]

[0088] This specific embodiment shows that the BP-ELM model can effectively identify 11 of the 12 complex faults, and the BPNN (BP neural network) can identify 9, with a success rate of 16.7%. In addition, although the model complexity is increased, the model training time is only increased by 0.02s.

[0089] The following combination Figure 7 , Figure 7 A structural block diagram of a fault diagnosis model training device provided by an embodiment of the present invention, the device may include:

[0090] A first training module 100 is used to extract a first feature of single fault data and a second feature of no-fault data, and input the first feature and the second feature into a preliminary diagnosis neural network for model training;

[0091] A first weight module 200 is used to extract a third feature of the composite fault data, and input the third feature into the trained preliminary diagnosis neural network to obtain a composite fault distribution weight;

[0092] A second weight module 300 is used to input the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight;

[0093] A second training module 400 is used to input the composite fault distribution weight and the single fault distribution weight into an ELM model for training;

[0094] The model building module 500 is used to connect the output layer of the trained preliminary diagnosis neural network with the input layer of the trained ELM model to obtain a fault diagnosis model.

[0095] Based on the above embodiments, the present invention avoids the problem of low diagnostic accuracy of the preliminary diagnostic neural network for composite fault detection when there are few composite fault samples by adding the ELM model for further detection.

[0096] Based on the above embodiments, the first training module 100 may include:

[0097] A first characteristic unit, wherein the first characteristic is obtained by processing single fault vibration data through wavelet packet decomposition and energy spectrum calculation;

[0098] A second feature unit, used for processing the fault-free vibration data by the wavelet packet decomposition and the energy spectrum calculation to obtain the second feature;

[0099] Accordingly, the first weight module 200 may include:

[0100] The third feature unit is used to process the composite fault vibration data through the wavelet packet decomposition and the energy spectrum calculation to obtain the third feature.

[0101] Based on the above embodiments, the wavelet packet decomposition is a four-layer wavelet packet decomposition.

[0102] Based on the above embodiments, the device may further include:

[0103] The sampling processing module is used to process the single fault data and the non-fault data by resampling and downsampling, so that the ratio of the data volume of each fault type in the single fault data to the non-fault data volume is equal.

[0104] Based on the above embodiments, the preliminary diagnosis neural network is a BP neural network.

[0105] The following combination Figure 8 , Figure 8 A structural block diagram of a fault diagnosis device provided by an embodiment of the present invention, the device may include:

[0106] An acquisition module 600 is used to acquire features of the data to be diagnosed;

[0107] The output module 700 is used to input the features into a fault diagnosis model to obtain a fault type diagnosis result; wherein the fault diagnosis model is a model trained by the fault diagnosis model training device.

[0108] The embodiment of the present invention avoids the problem of low accuracy of diagnostic results obtained by using the preliminary diagnostic neural network for composite fault detection when there are few composite fault samples by adding the ELM model for further detection.

[0109] Based on the above embodiments, the present invention further provides a fault diagnosis device, which may include a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps provided in the above embodiments may be implemented. Of course, the device may also include various necessary network interfaces, power supplies, and other components.

[0110] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by an execution terminal or a processor, the method provided in the embodiment of the present invention can be implemented; the storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0111] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0112] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0113] The above is a detailed introduction to a fault diagnosis model training method, a fault diagnosis method and related equipment provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A fault diagnosis model training method, It is characterized in that include: Extracting a first feature of single fault data and a second feature of no-fault data, and inputting the first feature and the second feature into a preliminary diagnosis neural network for model training; Extracting a third feature of the composite fault data, and inputting the third feature into the trained preliminary diagnosis neural network to obtain a composite fault distribution weight; Inputting the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight; Inputting the composite fault distribution weight and the single fault distribution weight into an ELM model for training; The output layer of the trained preliminary diagnosis neural network is connected to the input layer of the trained ELM model to obtain a fault diagnosis model.

2. According to the fault diagnosis model training method of claim 1, It is characterized in that The extracting of the first feature of the single fault data and the second feature of the no-fault data comprises: Processing the single fault vibration data by wavelet packet decomposition and energy spectrum calculation to obtain the first feature; Processing the fault-free vibration data by the wavelet packet decomposition and the energy spectrum calculation to obtain the second feature; Accordingly, the third feature of the composite fault data is extracted, including: The third feature is obtained by processing the composite fault vibration data through the wavelet packet decomposition and the energy spectrum calculation.

3. According to the fault diagnosis model training method of claim 2, It is characterized in that The wavelet packet decomposition is a four-layer wavelet packet decomposition.

4. According to the fault diagnosis model training method of claim 1, It is characterized in that Before extracting the first feature of the single fault data and the second feature of the no-fault data, the method further includes: The single fault data and the non-fault data are processed by resampling and downsampling, so that the ratio of the amount of each fault type data in the single fault data to the amount of non-fault data is equal.

5. According to the fault diagnosis model training method of claim 1, It is characterized in that The preliminary diagnosis neural network is a BP neural network.

6. A fault diagnosis method, It is characterized in that include: Acquire the characteristics of the data to be diagnosed; Inputting the features into a fault diagnosis model to obtain a fault type diagnosis result; Wherein, the fault diagnosis model is a model trained according to the fault diagnosis model training method according to any one of claims 1 to 5.

7. A fault diagnosis model training device, It is characterized in that include: A first training module is used to extract a first feature of single fault data and a second feature of no-fault data, and input the first feature and the second feature into a preliminary diagnosis neural network for model training; A first weight module is used to extract a third feature of the composite fault data, and input the third feature into the trained preliminary diagnosis neural network to obtain a composite fault distribution weight; A second weight module, used for inputting the first feature into the trained preliminary diagnosis neural network to obtain a single fault distribution weight; A second training module, used for inputting the composite fault distribution weight and the single fault distribution weight into an ELM model for training; The model building module is used to connect the output layer of the trained preliminary diagnosis neural network with the input layer of the trained ELM model to obtain a fault diagnosis model.

8. A fault diagnosis device, It is characterized in that include: An acquisition module, used to acquire the characteristics of the data to be diagnosed; An output module, used for inputting the features into a fault diagnosis model to obtain a fault type diagnosis result; Wherein, the fault diagnosis model is a model trained by the fault diagnosis model training device according to claim 7.

9. A fault diagnosis device, It is characterized in that include: Memory, for storing computer programs; A processor, used to implement the fault diagnosis model training method as described in any one of claims 1 to 5 and / or the fault diagnosis method as described in claim 6 when executing the computer program.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by the processor, the fault diagnosis model training method according to any one of claims 1 to 5 and / or the fault diagnosis method according to claim 6 are implemented.

Citation Information

Patent Citations

  • Transformer fault detection method and related device

    CN110516757A

  • Bearing fault diagnosis method and system based on improved BP neural network

    CN113465924A